From Adoption to Trust in AI-Assisted Environmental Evidence Synthesis: A Systematic Review of Capabilities, Reliability Boundaries, and Real-World Applications

Environmental evidence synthesis (EES) transforms diverse environmental evidence into trustworthy and decision-relevant knowledge for environmental governance and policymaking. Yet, the accelerating growth and heterogeneity of environmental literature are increasingly exceeding the capacity of conventional manual workflows. Here, we conducted an expert-led systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidance to assess the capabilities, reliability boundaries, and real-world applications of artificial intelligence (AI)-assisted EESs. From 38,840 records, we included 459 peer-reviewed studies and organized them into three evidence streams. Capability evidence revealed a substantial but weakly EES-adapted AI tool ecosystem with 158 directly usable tools but only three directly relevant to EES, indicating that environment-specific capabilities remain scarce and task-localized. Reliability evidence showed task-contingent AI performance, with more stable outputs for structured front-end tasks but persistent verification needs for judgment-intensive tasks. Real-world evidence shows that AI has already entered evidence synthesis practice, but its use in EES is advancing ahead of domain-specific validation. Together, these findings indicate that AI-assisted EES is moving from possibility to practice but not yet from adoption to trust. This transition will require trustworthy AI-assisted EES infrastructure, including task-specific environmental tools, expert-annotated benchmarks, transparent reporting and validation standards, and living EES systems for continuous updates. Only within such an infrastructure can AI move beyond a tool for efficiency gains and become a governed, credible capability for environmental knowledge production.